A chip multiprocessor die supports optional stacking of additional dies. The chip multiprocessor includes a plurality of processor cores, a memory controller, and stacked cache interface circuitry. The stacked cache interface circuitry is configured to attempt to retrieve data from a stacked cache die if the stacked cache die is present but not if the stacked cache die is absent. In one implementation, the chip multiprocessor die includes a first set of connection pads for electrically connecting to a die package and a second set of connection pads for communicatively connecting to the stacked cache die if the stacked cache die is present. Other embodiments, aspects and features are also disclosed.
|
17. A packaged chip comprising:
a processor die including at least one processor core and a circuit;
a first cache die separate from the processor die and having a cache memory,
wherein the processor die and first cache die are at different stacking levels of a stacked arrangement of plural dies, and
wherein the circuit is to detect whether the cache die is present in the packaged chip.
1. A chip multiprocessor die supporting optional stacking of additional dies, the chip multiprocessor die comprising:
a plurality of processor cores;
a memory controller; and
stacked cache interface circuitry which is configured to attempt to retrieve data from a stacked cache die separate from the chip multiprocessor die if the stacked cache die is present but not to attempt to retrieve data from the stacked cache die if the stacked cache die is absent,
wherein the chip multiprocessor die is to be provided in a stacked arrangement of dies at a stacking level different from a stacking level of the stacked cache die that is also part of the stacked arrangement of dies.
9. A packaged chip multiprocessor comprising:
a chip multiprocessor die;
a plurality of processor cores on the chip multiprocessor die;
a memory controller on the chip multiprocessor die;
a first set of connection pads on the chip multiprocessor die which are electrically connected to a package; and
a second set of connection pads on the chip multiprocessor die which are configured to be connectable to a stacked cache die if present, wherein the stacked cache die is separate from the chip multiprocessor die, and wherein the chip multiprocessor die is to be provided in a stacked arrangement of dies at a stacking level different from a stacking level of the stacked cache die; and
stacked cache interface circuitry which is configured to attempt to retrieve data from the stacked cache die if present.
2. The chip multiprocessor die of
3. The chip multiprocessor die of
4. The chip multiprocessor die of
5. The chip multiprocessor die of
6. The chip multiprocessor die of
7. The chip multiprocessor die of
8. The chip multiprocessor die of
10. The packaged chip multiprocessor of
at least one cache memory on the chip multiprocessor die, and
the stacked cache die, wherein the stacked cache die and the chip multiprocessor die form the stacked arrangement of dies in which the chip multiprocessor die is at a first stacking level and the stacked cache die is at a second, different stacking level, wherein the stacked cache die includes a cache memory at a cache level different from a cache level of the at least one cache memory on the chip multiprocessor die.
11. The packaged chip multiprocessor of
12. The packaged chip multiprocessor of
13. The packaged chip multiprocessor of
14. The packaged chip multiprocessor of
15. The packaged chip multiprocessor of
16. The packaged chip multiprocessor of
the stacked cache die, wherein the stacked cache die and the chip multiprocessor die are part of the stacked arrangement of dies.
18. The packaged chip of
19. The packaged chip of
20. The packaged chip of
|
1. Technical Field
The present application relates generally to processors and memory for computer systems.
2. Description of the Background Art
Conventional two-dimensional (2-D) microprocessors, including conventional chip multiprocessors, are formed on a single silicon die. In order to increase performance of these microprocessors, further components, such as more processor cores, caches and memory controllers, are generally being integrated into the single silicon die.
Recently, however, technologies for stacking of silicon die have been developed. In order to apply the stacking technologies to chip multiprocessors, various proposals have been made. Each of these proposals provide an architecture or design for implementing the chip multiprocessor on a stack of silicon dies. For example, one set of proposals splits each core of the chip multiprocessor between multiple stacked die.
Applicants have observed that each of the proposals for applying stacking to chip multiprocessors makes the natural assumption that stacking will be required. In other words, the designs are optimized assuming stacking of silicon dies.
One embodiment relates to a chip multiprocessor die supporting optional stacking of additional dies. The chip multiprocessor includes a plurality of processor cores, a memory controller, and stacked cache interface circuitry. The stacked cache interface circuitry is configured to attempt to retrieve data from a stacked cache die if the stacked cache die is present but not if the stacked cache die is absent. In one implementation, the chip multiprocessor die includes a first set of connection pads for electrically connecting to a die package and a second set of connection pads which can be configured for communicatively connecting to the stacked cache die if the stacked cache die is present.
Other embodiments, aspects, and features are also disclosed.
The present application discloses an architectural design for a chip multiprocessor die in embodiments of the present invention, where the chip microprocessor die is configured to be modular in that the 3-D stacking of additional levels of cache memory is optional, i.e. possible but not required. In this architecture, all the cores are contained on a single processor die. Additional cache levels may be optionally stacked using additional die.
The heat sink 104 may be advantageously attached to the chip multiprocessor die 102, and the package 106 (including connectors 108 for power and input/output) is preferably attached to at least one stacked cache die (for example, the topmost stacked die 110-2 in the illustrated example), if any. Of course, while the CMP 102, stacked cache die 110, and package 106 are shown spaced apart in
As shown in
As shown in
Advantageously, using this architectural design, the number of stacked stacked cache die 110 is variable. In the particular implementation shown in
Furthermore, this architectural design advantageously positions the cores, which typically dissipate the vast majority of power and hence generate the most heat, nearest to the heat sink 104 and the optional stacked cache die, which typically generate much less heat, further from the heat sink 104.
An example logical design for a chip multiprocessor 102 in accordance with an embodiment of the present invention is illustrated in
As shown in
As further shown in
The stacked cache interface circuitry 312 may be small and so may be implemented without adding much cost to the CMP die 102 in the case where the CMP die 102 is not stacked (i.e. where no stacked cache die 110 are used and there are no stack die connections 114). Also, power to the stacked cache interface circuitry 312 may be configured so as to be unconnected in the case where the CMP die 102 is not stacked.
The memory controllers 310 may be configured to signal the stacked cache interface circuitry 312 so as to find out if one or more stacked cache die are present or if there are no stacked cache die. The stacked cache interface circuitry 312 may be configured to detect the presence of the optional stacked cache (e.g., a level 4 cache) die 110 by several mechanisms. One such mechanism comprises receiving a reply (acknowledgement signal) to signals transmitted to the stacked cache die 110 to indicate presence of the stacked cache die 110 or not receiving a reply to such signals which would indicate an absence of the stacked cache die 110. Another mechanism comprises an absence of a signal path due to a lack of stacking (i.e. the signal path is open circuit when there is no stacked cache die 110).
The memory controller 310 receives 402 a memory request. If no stacked cache die is present, then the memory controller 310 fetches 406 the requested data from the main memory (for example, from the memory DIMMs). In other words, in the lower-performance lower-cost configuration shown in
On the other hand, if there is a stacked cache (i.e. if there is one or more cache die) 110, then the memory controller 310 attempts to find the data in the stacked cache 110 by sending 408 a memory request signal to the stacked cache interface circuitry 312. If 410 the data is found in the stacked cache (i.e. a stacked cache hit), then the memory controller 310 receives 412 the requested data from the stacked cache interface 312. If the data cannot be found in the stacked cache (i.e., a stacked cache miss), then the memory controller 310 resorts to fetching 406 the data from memory.
In accordance with one embodiment, the chip multiprocessor die 102 includes one or more near cache levels, and the stacked cache die(s) 110 include optional cache levels which are higher (farther) than those levels on the chip multiprocessor die 102. In that case, memory requests would be processed by first checking the near cache levels on the CMP die 102. Upon near cache misses such that the data requested is not found on the CMP die 102, the memory controller 310 would then send a memory request signal to the stacked cache interface circuitry 312. If the data is found in the stacked cache (i.e. a stacked cache hit), then the memory controller 310 receives 412 the requested data from the stacked cache interface 312. If 410 the data cannot be found in the stacked cache (i.e., a stacked cache miss), then the memory controller 310 resorts to fetching 406 the data from memory.
The architecture disclosed in the present application has several advantages or potential advantages. First, processor manufacturers are generally interested in providing a range of products to cover different markets. However, designing different products for different markets is typically rather expensive. Instead, with optional stacking, manufacturers may sell a stacked chip microprocessor in markets that had higher performance demands and required a more powerful memory system (e.g., enterprise servers and high-performance computing applications), while the base die may be used in lower-performance cost-sensitive applications (e.g., home consumer and laptop applications). The lower-performance lower-cost version of the product may be built simply by omitting some or all of the stacked die.
Second, by placing all the cores on a single die, this architecture produces a low thermal resistance between the cores and the heat sink. Since the cores dissipate the vast majority of the power, this yields the lowest operating temperature (i.e. most efficient heat sinking) for the stack as a whole.
In the above description, numerous specific details are given to provide a thorough understanding of embodiments of the invention. However, the above description of illustrated embodiments of the invention is not intended to be exhaustive or to limit the invention to the precise forms disclosed. One skilled in the relevant art will recognize that the invention can be practiced without one or more of the specific details, or with other methods, components, etc. In other instances, well-known structures or operations are not shown or described in detail to avoid obscuring aspects of the invention. While specific embodiments of, and examples for, the invention are described herein for illustrative purposes, various equivalent modifications are possible within the scope of the invention, as those skilled in the relevant art will recognize.
These modifications can be made to the invention in light of the above detailed description. The terms used in the following claims should not be construed to limit the invention to the specific embodiments disclosed in the specification and the claims. Rather, the scope of the invention is to be determined by the following claims, which are to be construed in accordance with established doctrines of claim interpretation.
Patent | Priority | Assignee | Title |
10381067, | Sep 26 2005 | Rambus Inc. | Memory system topologies including a buffer device and an integrated circuit memory device |
10535398, | Sep 26 2005 | Rambus Inc. | Memory system topologies including a buffer device and an integrated circuit memory device |
10672458, | Sep 26 2005 | Rambus Inc. | Memory system topologies including a buffer device and an integrated circuit memory device |
11043258, | Sep 26 2005 | Rambus Inc. | Memory system topologies including a memory die stack |
11328764, | Sep 26 2005 | Rambus Inc. | Memory system topologies including a memory die stack |
11727982, | Sep 26 2005 | Rambus Inc. | Memory system topologies including a memory die stack |
8105882, | Feb 14 2007 | Hewlett Packard Enterprise Development LP | Processing a memory request in a chip multiprocessor having a stacked arrangement |
8386690, | Nov 13 2009 | International Business Machines Corporation | On-chip networks for flexible three-dimensional chip integration |
8799710, | Jun 28 2012 | GLOBALFOUNDRIES U S INC | 3-D stacked multiprocessor structures and methods to enable reliable operation of processors at speeds above specified limits |
8826073, | Jun 28 2012 | GLOBALFOUNDRIES U S INC | 3-D stacked multiprocessor structures and methods to enable reliable operation of processors at speeds above specified limits |
8987066, | Jan 03 2012 | Honeywell International Inc. | Processing unit comprising integrated circuits including a common configuration of electrical interconnects |
9190118, | Nov 09 2012 | GLOBALFOUNDRIES Inc | Memory architectures having wiring structures that enable different access patterns in multiple dimensions |
9195630, | Mar 13 2013 | International Business Machines Corporation | Three-dimensional computer processor systems having multiple local power and cooling layers and a global interconnection structure |
9257152, | Nov 09 2012 | GLOBALFOUNDRIES Inc | Memory architectures having wiring structures that enable different access patterns in multiple dimensions |
9298672, | Apr 20 2012 | International Business Machines Corporation | 3-D stacked multiprocessor structure with vertically aligned identical layout operating processors in independent mode or in sharing mode running faster components |
9336144, | Jul 25 2013 | GLOBALFOUNDRIES U S INC | Three-dimensional processing system having multiple caches that can be partitioned, conjoined, and managed according to more than one set of rules and/or configurations |
9383411, | Jun 26 2013 | International Business Machines Corporation | Three-dimensional processing system having at least one layer with circuitry dedicated to scan testing and system state checkpointing of other system layers |
9389876, | Oct 24 2013 | International Business Machines Corporation | Three-dimensional processing system having independent calibration and statistical collection layer |
9391047, | Apr 20 2012 | International Business Machines Corporation | 3-D stacked and aligned processors forming a logical processor with power modes controlled by respective set of configuration parameters |
9412718, | Apr 20 2012 | International Business Machines Corporation | 3-D stacked and aligned processors forming a logical processor with power modes controlled by respective set of configuration parameters |
9442884, | Apr 20 2012 | International Business Machines Corporation | 3-D stacked multiprocessor structures and methods for multimodal operation of same |
9471535, | Apr 20 2012 | International Business Machines Corporation | 3-D stacked multiprocessor structures and methods for multimodal operation of same |
9569402, | Apr 20 2012 | International Business Machines Corporation | 3-D stacked multiprocessor structure with vertically aligned identical layout operating processors in independent mode or in sharing mode running faster components |
9696379, | Jun 26 2013 | International Business Machines Corporation | Three-dimensional processing system having at least one layer with circuitry dedicated to scan testing and system state checkpointing of other system layers |
Patent | Priority | Assignee | Title |
5649144, | Jun 13 1994 | HEWLETT-PACKARD DEVELOPMENT COMPANY, L P | Apparatus, systems and methods for improving data cache hit rates |
5828578, | Nov 29 1995 | SAMSUNG ELECTRONICS CO , LTD | Microprocessor with a large cache shared by redundant CPUs for increasing manufacturing yield |
6247094, | Dec 22 1997 | Intel Corporation | Cache memory architecture with on-chip tag array and off-chip data array |
6349360, | Nov 09 1999 | International Business Machines Corporation | System bus read address operations with data ordering preference hint bits |
6647463, | Sep 08 2000 | NEC Corporation | Cache update method and cache update control system employing non-blocking type cache |
6725334, | Jun 09 2000 | SK HYNIX INC | Method and system for exclusive two-level caching in a chip-multiprocessor |
6848071, | Apr 23 2001 | Oracle America, Inc | Method and apparatus for updating an error-correcting code during a partial line store |
6972494, | Aug 17 1999 | Bell Semiconductor, LLC | Integrated circuit die for wire bonding and flip-chip mounting |
6976131, | Aug 23 2002 | Sony Corporation of America | Method and apparatus for shared cache coherency for a chip multiprocessor or multiprocessor system |
7305522, | Feb 12 2005 | International Business Machines Corporation | Victim cache using direct intervention |
Executed on | Assignor | Assignee | Conveyance | Frame | Reel | Doc |
Feb 09 2007 | JOUPPI, NORMAN PAUL | HEWLETT-PACKARD DEVELOPMENT COMPANY, L P | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 019038 | /0545 | |
Feb 14 2007 | Hewlett-Packard Development Company, L.P. | (assignment on the face of the patent) | / | |||
Oct 27 2015 | HEWLETT-PACKARD DEVELOPMENT COMPANY, L P | Hewlett Packard Enterprise Development LP | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 037079 | /0001 |
Date | Maintenance Fee Events |
Mar 08 2013 | M1551: Payment of Maintenance Fee, 4th Year, Large Entity. |
Apr 21 2017 | M1552: Payment of Maintenance Fee, 8th Year, Large Entity. |
Apr 22 2021 | M1553: Payment of Maintenance Fee, 12th Year, Large Entity. |
Date | Maintenance Schedule |
Nov 10 2012 | 4 years fee payment window open |
May 10 2013 | 6 months grace period start (w surcharge) |
Nov 10 2013 | patent expiry (for year 4) |
Nov 10 2015 | 2 years to revive unintentionally abandoned end. (for year 4) |
Nov 10 2016 | 8 years fee payment window open |
May 10 2017 | 6 months grace period start (w surcharge) |
Nov 10 2017 | patent expiry (for year 8) |
Nov 10 2019 | 2 years to revive unintentionally abandoned end. (for year 8) |
Nov 10 2020 | 12 years fee payment window open |
May 10 2021 | 6 months grace period start (w surcharge) |
Nov 10 2021 | patent expiry (for year 12) |
Nov 10 2023 | 2 years to revive unintentionally abandoned end. (for year 12) |